This is the mail archive of the
libc-alpha@sourceware.org
mailing list for the glibc project.
Re: Benchmark for sem_timedwait
- From: Siddhesh Poyarekar <siddhesh at redhat dot com>
- To: OndÅej BÃlka <neleai at seznam dot cz>
- Cc: Roland McGrath <roland at hack dot frob dot com>, libc-alpha at sourceware dot org
- Date: Thu, 13 Feb 2014 15:52:00 +0530
- Subject: Re: Benchmark for sem_timedwait
- Authentication-results: sourceware.org; auth=none
- References: <20140211172324 dot GA31923 at domone dot podge> <20140211172720 dot 52ABB7445F at topped-with-meat dot com> <20140211184530 dot GB32451 at domone dot podge> <20140211204751 dot GA17236 at domone dot podge>
On Tue, Feb 11, 2014 at 09:47:51PM +0100, OndÅej BÃlka wrote:
> I used a same test for sem_timedwait. Results look inconclusive for
> uncontended case and in contended case assembly is bit faster.
>
> There is again question if assembly implementation is worth it, or if it
> suffices to write only hot path in assembly and refactor a cold
> implementation to have a futex path in separate function that would be
> called from assembly.
>
> Comments on this?
Please try to use bench-skeleton.c. I think Torvald had proposed some
patches a while back that should enable you to do this, although the
actual benchmarks to use those patches didn't make it in AFAICT.
If using bench-skeleton.c is not possible directly, I would appreciate
improvements to it to actually make it work.
Siddhesh